Cao Huu Hiep

AI Engineer

https://talent.gravityer.com/cao-huu-hiep

AI Engineer with less than a year in AI pipeline development and data engineering.

Key Strengths

Strong foundational knowledge in Deep Learning architectures (Transformers, CNN, RNN) and frameworks (PyTorch, TensorFlow, Hugging Face).
Demonstrated ability to design and deploy end-to-end AI pipelines, including data preprocessing, model inference, and post-processing.
Extensive experience with Retrieval Augmented Generation (RAG) systems, including multi-stage retrieval, RRF, and cross-encoder rerankers.
Proficiency in building scalable data pipelines for various AI tasks, including dataset generation, edge-case analysis, and automated evaluation.
Experience with real-time AI system design, utilizing multi-threading, queue-based buffering, and optimized CPU/GPU utilization.
Practical experience with fine-tuning large models (e.g., Whisper with QLORA) and performance optimization.
Exposure to geospatial data engineering and 3D visualization, indicating versatility beyond core NLP/ML.
Solid understanding of evaluation metrics for NLP tasks (Recall@5, WER, CER, ROUGE, F1-score).

Cultural & Operational Fit

Cultural Fit Analysis

The candidate's academic projects (DUT-RAG, SpeechBridge, Vietnamese Abstractive Summarization) demonstrate a strong interest and practical application in AI/ML, aligning well with an AI Engineer role. The internship experiences at VJ Technologies and MakeAI further solidify this alignment, showcasing real-world application of AI principles and system development. The diversity of projects, from RAG systems to speech translation and geospatial AI, indicates a broad skill set and adaptability, which are positive indicators for cultural fit in a dynamic AI environment. The focus on performance optimization and scalable solutions suggests a results-oriented mindset. The candidate is currently pursuing a Bachelor's degree, which might indicate a preference for roles that support continued learning and growth.

Soft Skills & Operational Fit

The candidate's project descriptions highlight strong problem-solving skills, particularly in optimizing complex AI systems for performance and recall. The individual project on Vietnamese Abstractive Summarization demonstrates initiative and independent work capability. The collaborative nature of the RAG and Speech Translation projects suggests good teamwork potential. The detailed descriptions of technical challenges and solutions indicate a methodical and analytical approach to engineering. However, without direct interview data, assessing communication clarity, stress handling, and team collaboration in a live operational setting is limited.

AI is analyzing your overall score…

Identifying your key strengths…

Evaluating your skill match against the job requirements…

Assessing your cultural and operational fit

Projects

Institutional Knowledge RAG System (DUT-RAG)

March 1, 2026 – May 1, 2026

Designed and deployed a full-stack RAG system processing 1,000+ multi-domain institutional documents (legal files & web news) utilizing automated web crawlers and OCR pipelines. Implemented a multi-stage retrieval architecture combining Sparse (BM25) and Dense Embeddings, integrated with Reciprocal Rank Fusion (RRF) and cross-encoder Rerankers (Cohere, local BGE), achieving an exceptional 95-100% Recall@5. Deployed a high-performance system serving 16,400+ data chunks, delivering 95-100% retrieval recall, <50ms search response time, and <1.5s LLM generation latency.

SpeechBridge - End-to-end Speech Translation Pipeline

February 1, 2026 – March 1, 2026

Designed and developed an end-to-end English-to-Vietnamese speech translation system integrating Whisper (ASR), M2M100 (NMT), and SpeechT5 (TTS) for seamless cross-lingual communication. Fine-tuned Whisper-small using 4-bit QLORA on cloud GPUs, enabling parameter-efficient training while maintaining high transcription quality. Built a scalable parallel audio preprocessing pipeline with VTT parsing and ±150ms boundary padding to improve speech-text alignment and dataset quality. Achieved a 10.5% relative reduction in both Word Error Rate (WER) and Character Error Rate (CER) compared to the baseline Whisper-small model.

Vietnamese Abstractive Summarization

September 1, 2025 – December 1, 2025

Developed an end-to-end Vietnamese abstractive summarization system using a custom Transformer architecture in PyTorch for automated news article summarization. Built a scalable data pipeline to collect and process 438K+ samples from the VietNews-Abs-Sum dataset, including custom Vietnamese BPE tokenization and Unicode normalization. Implemented encoder-decoder Transformer with multi-head attention and positional encoding, evaluated using ROUGE, F1-score, and accuracy metrics.

Key Strengths

Strong foundational knowledge in Deep Learning architectures (Transformers, CNN, RNN) and frameworks (PyTorch, TensorFlow, Hugging Face).
Demonstrated ability to design and deploy end-to-end AI pipelines, including data preprocessing, model inference, and post-processing.
Extensive experience with Retrieval Augmented Generation (RAG) systems, including multi-stage retrieval, RRF, and cross-encoder rerankers.
Proficiency in building scalable data pipelines for various AI tasks, including dataset generation, edge-case analysis, and automated evaluation.
Experience with real-time AI system design, utilizing multi-threading, queue-based buffering, and optimized CPU/GPU utilization.
Practical experience with fine-tuning large models (e.g., Whisper with QLORA) and performance optimization.
Exposure to geospatial data engineering and 3D visualization, indicating versatility beyond core NLP/ML.
Solid understanding of evaluation metrics for NLP tasks (Recall@5, WER, CER, ROUGE, F1-score).

Cultural & Operational Fit

Cultural Fit Analysis

Soft Skills & Operational Fit

Cao Huu Hiep

Key Strengths

Cultural & Operational Fit

About

Top Skills

Skills

Education

Experience

Projects

Key Strengths

Cultural & Operational Fit